Research On The Predicting Model Of Multifactor Time Series Based On Feature Selection

There is a great deal of nonlinear, complex time-series of high-dimensional features in natural and socio-economic phenomenon. The prediction of these time series acting as a basis for decision-making is of great realistic significance. In recent years, with the mutual penetration and influence in all spheres of the society, leading to the dimension of multifactor time series is increasing fiercely. However, the traditional method of statistical analysis in dealing with nonlinear problems has many defects. When dealing with the time series of high-dimensional features, traditional neural networks have not been satisfactory in both efficiency and results. The reason is that the time series of high-dimensional features implied a lot of irrelevant and redundant information, and it reduces the precision and efficiency of the neural network. So feature selection about the time series of high-dimensional feature is very necessary. In this dissertation, combineing feature selection and RBF neural network for time series prediction together, and multifactor time series predicting model based on PRN feature selection algorithm is proposed to improve the efficiency, accuracy and generalization of prediction.The content of this dissertation is as follows:(1)It researches the better recognized relief algorithm in feature evaluation, analyzes its disadvantages, improves it using the combination of vertical and horizontal two-way compression redundant data and learning algorithm, and a modified combined feature selection algorithm is proposed. and analyzes its application of classification and regression.(2)It analyzes the characteristics of multifactor time series data, proposes that RreliefF algorithm which is generally used in regression should be used for the feature selection of multifactor time series. PRN combined feature selection algorithm is proposed.(3)On the basis of the research of RBF neural network for time series prediction and PRN combined feature selection algorithm, the predicting model of multifactor time series based on PRN combined feature selection algorithm is established.